Pattern Recognition (PR) & Neural Networks (NN)

Transcription

1 Pattern Recognition (PR) & Neural Networks (NN) by Pascual Campoy Computer Vision Group Universidad Politécnica Madrid Guide to the subject Objectives Methodology Learning materials Evaluation Topics Schedule 2

2 Objectives: Be able to explain following terms and relate them to real-word problems: - learning, pattern recognition, classification, dimensionality reduction, supervised learning, unsupervised learning Ability to apply classical PR techniques - PCA, Bayesian Decision & K-means Ability to apply supervised ANN: - MLP to PR Ability to apply unsupervised ANN: - SOM to PR 3 SCALE-UP Methodology classroom at MIT 4

3 Methodology In the classroom: - Lecture - Colaborative working on the computer - Tutorial - Presentation of practical works Out of the classroom: - Individual study (bofore-after) - 2 Practical Works 5 Learning material Aulaweb: invited student in 70038, password aprendizaje - This guide: 0_guide_PR_NN.pdf - Dlides for every topic, including classroom exercices - Dataset for exercises and pracical works, including exercise form _plantilla_ejercicios_clase.doc - 2 Practical work statement 6

4 Learning material "Pattern Classification" Duda-R, Hart-P, Stork-D Wiley-Interscience, 2004 Neural Networks for Pattern recognition Christopher M. Bishop Oxford Press, Further reading biological inspiration Christof Koch Rodolfo Llinás V.S. Ramachandran 8

5 further reading making things to work Jeff Hawkins David Fogel 9 Evaluation Continuous evaluation - Class-room exercises - Practical works - Exam (min 5/10) Momentary evaluation - Practical works (compulsatory) - Exam (min 5/10) 1,5 8,5 10

6 Topics: 1. Intelligence: PR & learning 2. Classical techniques 3. Learning methodology: ANN 4. Supervised ANN: multilayer perceptron 5. Non supervised ANN: Self-organized maps 6. Research & challenges 11 Schedule (1/2) Schedule for "Pattern Recognition & Neural Networks" Topic Subject Week/day 0 Guide to the subject 30-S 1 Introduction: Intelligence, Learning & Pattern Recognition 7-O 2 Classical techniques: PCA 14-O, 21-O 2 Classical techniques: Classifiers 28-O 3 Machine learning and Neural Networks 4-N 4 Supervised Neural Networks: MLP 11-N 18-N 1-2 Presentation Parctical Work #1 25-N 5 Unsupervised Neural Networks: SOM 2-D, 9-D Review 16-D X-mas rest 4-5 Presentation Parctical Work #2 13-E 6 "State of the art & research" 20-E Exam 26-E 12

7 Schedule (2/2) 3 ECTS x 25 hours/ects = 75 hours - Classroom: 28 hours = 14 weeks x 2 hours/week - Outside the classroom 47 hours: 18.5 hours for studying hours for practical works (2,7 hours/week) 10 hours for preparing final exam 13 Classroom exercices 1. Create a word document from the template plantilla_ejercicios_clase.doc (downloaded from Aulaweb/contenidos/problemas) with the name: ex_yy_aaaaa_bbbbb_ccccc.doc where X is the chapter number YY is the exercise number within the chapter AAAAA, BBBBB and CCCCC are the students ID numbers 2. Save this document in your PC at Documentos compartidos/entregar 14

8 classroom exercices 3. Fullfill the heading: 4. Write in the document the required solution that includes the explanation, the Matlab code, the obtained graphics and results, and the comments and conclusions 5. The document has to be closed in order to allow to be collected 15 Example 0.1 change Matlab working directory to Documentos_ compartidos/entregar download datos_d2_c3_s1 from Aulaweb into this directory >> load datos_d2_c3_s1 p.valor 2x1000 p.clase 1x1000 p.salida 1x1000 >> plot(p.valor(1,:),p.valor(2,:), b. ); hold on; 16

9 Exercise 0.1 >> load datos_d2_c3_s1.mat Print p.valor using a diferent color/prompt for each clase determined in p.clase 17 What is PR by learning?? 19